A ChatGPT alternative that checks its own translation

Because the translating is not the hard part any more; checking it is. ChatGPT will happily hand you 400 translated strings, and nothing in that answer tells you which three drifted in meaning or lost a placeholder. KAERIS translates every line, then translates it back into your language with a different model and shows you what each line now says. In our benchmark that caught 28 of 28 planted meaning errors with 0 false positives across 40 heavily reworded paraphrases.

For a handful of strings, yes — and we would not pretend otherwise. It stops scaling at about the point where the file no longer fits in one message: you chunk it, keys get renamed or dropped between chunks, nesting flattens, plural forms come back in the wrong shape, and you re-paste your glossary in every new chat. KAERIS takes the file, writes the same format back, and keeps keys, nesting and plural rules as a guarantee rather than a hope.

It uses commercial models through OpenRouter, and deliberately not the same one twice: the model that translates a line is never the model that reads it back. A model grading its own work agrees with itself. On the Lifetime plan you bring your own API key and choose the provider yourself.

KAERIS trains no models — we don't operate any. Your file is deleted from our server within 2 hours; the translation runs through OpenRouter to the model providers under their own policies. For full control, Lifetime is BYOK (your own key). More on security.

Everything that is a conversation: deciding on the wording in the first place, writing the copy, arguing about tone, explaining a locale convention you have never met. KAERIS does not do any of that. It is a pipeline for files, not an assistant.